INTELLIGENT CLASSIFICATION OF HEALTHCARE INCIDENT REPORTS USING SUPERVISED LEARNING

The reporting of incidents in healthcare is crucial for enhancing patient safety and safeguarding healthcare personnel. But the classification of incident reports manually is time-consuming, inconsistent and prone to human error, which requires automated solutions. The primary objective of this study was to create an automatic Healthcare Incident Classification system based on the NLP and machine learning techniques to improve patient safety, and to monitor cybersecurity. The incident narratives were tokenized, lemmatized, the stop words were removed and the Term Frequency–Inverse Document Frequency (TF-IDF) was vectorised. A total of 209 incident reports were collected from a private hospital in Ibadan, Nigeria and these were used to train multiple machine learning models (XGBoost, SVM, Random Forest, Naïve Bayes, Logistic Regression and K-Nearest Neighbor). The dataset was split into training and testing sets in a 70:30 ratio, and model performance was evaluated using accuracy, precision, recall, and F1-score. Among these several machine learning models. XGBoost achieved the best performance with 69% accuracy, 100% precision, 93% recall, and 97% F1-score for critical severity classification. Training on augmented datasets (750, 3,056, and 9,000 samples) further improved performance, achieving up to 69% accuracy and 100% precision, recall, and F1-score for the sample of 9000. The suggested system can be integrated into the hospital system to facilitate prompt detection of incidents and improve patient safety